output

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7883
• Wer: 1.0199

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.92	0.95	400	2.9522	1.0026
1.0435	1.89	800	0.8608	1.0552
0.5354	2.84	1200	0.7762	1.0169
0.404	3.79	1600	0.6984	1.0293
0.3301	4.73	2000	0.6811	1.0217
0.2745	5.68	2400	0.7027	1.0308
0.2346	6.63	2800	0.7296	1.0185
0.2096	7.57	3200	0.7148	1.0294
0.1912	8.52	3600	0.7109	1.0335
0.172	9.47	4000	0.7894	1.0252
0.1567	10.41	4400	0.7592	1.0219
0.1457	11.36	4800	0.8030	1.0141
0.1337	12.31	5200	0.7811	1.0237
0.1288	13.25	5600	0.7703	1.0188
0.1165	14.2	6000	0.7728	1.0199
0.105	15.15	6400	0.7934	1.0206
0.1028	16.09	6800	0.7978	1.0185
0.092	17.04	7200	0.8276	1.0289
0.0901	17.99	7600	0.7881	1.0202
0.0818	18.93	8000	0.7847	1.0162
0.0801	19.88	8400	0.8142	1.0230
0.0768	20.83	8800	0.7735	1.0215
0.0721	21.78	9200	0.7941	1.0227
0.0658	22.72	9600	0.8100	1.0219
0.0627	23.67	10000	0.7592	1.0196
0.0591	24.62	10400	0.8028	1.0210
0.0537	25.56	10800	0.8019	1.0253
0.0507	26.51	11200	0.7951	1.0212
0.0495	27.46	11600	0.7893	1.0207
0.0466	28.4	12000	0.7854	1.0188
0.0431	29.35	12400	0.7883	1.0199

Framework versions

• Transformers 4.32.1
• Pytorch 2.2.0+cu121
• Datasets 2.12.0
• Tokenizers 0.13.2